[PATCH] sym2: Use ORDERED_QUEUE_TAG when barrier request is issued

[PATCH] sym2: Use ORDERED_QUEUE_TAG when barrier request is issued

[Hifumi Hisashi]

Hi.
I think using ORDERED_QUEUE_TAG is preferable instead of
SIMPLE_QUEUE_TAG when a request with barrier flag is
issued.

Thanks.

Signed-off-by: Hifumi Hisashi <hifumi.hisashi@lab.ntt.co.jp>

--- linux-2.6.16-rc3/drivers/scsi/sym53c8xx_2/sym_hipd.c	2006-02-13 
14:04:02.000000000 +0900
+++ linux-2.6.16-rc3_fix/drivers/scsi/sym53c8xx_2/sym_hipd.c	2006-02-13 
14:36:47.000000000 +0900
@@ -41,6 +41,7 @@
  #include <linux/slab.h>
  #include <asm/param.h>		/* for timeouts in units of HZ */
  #include <scsi/scsi_dbg.h>
+#include <scsi/scsi_tcq.h>

  #include "sym_glue.h"
  #include "sym_nvram.h"
@@ -5109,6 +5110,9 @@
  			lp->tags_since = 0;
  		}
  #endif
+		if (cmd->request && cmd->request->flags & REQ_HARDBARRIER)
+                        order = M_ORDERED_TAG;
+
  		msgptr[msglen++] = order;

  		/* 

Re: [PATCH] sym2: Use ORDERED_QUEUE_TAG when barrier request is issued

[Matthew Wilcox]

On Tue, Feb 14, 2006 at 01:10:54AM +0900, Hifumi Hisashi wrote:
> I think using ORDERED_QUEUE_TAG is preferable instead of
> SIMPLE_QUEUE_TAG when a request with barrier flag is
> issued.

This isn't enough; we don't know if the target device supports ordered
tags or not.  The right way to do this is to convert sym2 to use the
midlayer tag code.  I think it's nearly at a point where the conversion
is feasible.  I'm going to push out the current set of cleanups next
week and then look at the problem again.